Large scale graph mining for web reputation inference

摘要: The explosion of the number devices and users on Internet results in massive amounts data information. This poses most complex challenges security we have ever faced. detection malicious domains protocol (IP) addresses has been a hot topic cyber security. We present scalable effective graph inference system for detecting IP addresses. goal is to protect from network threats. Based loopy belief propagation algorithm, infers every domain or reputation, flagging it with high reputation as one. evaluated 75 million-node constructed huge dataset (500 gigabytes). attains performance 86% 87% area under receiver operating curves inferring reputations respectively. demonstrate that graphical solution provides rapid assessment safe risky sites McAfee's data. It an automatic tool web field serves assisting “first pass” classification triaging.